Latest / Today’s AI News / Musk vs Altman in Federal Court, Google Signs Pentagon Deal, an AI That Thinks It's 1930

Transcript

Musk vs Altman in Federal Court, Google Signs Pentagon Deal, an AI That Thinks It's 1930 · Today’s AI News

• 0:00
  Welcome back to today's AI News, brought to you by NYNEX
• 0:02
  Productions. Imagine 2 brilliant architects
• 0:05
  who build a world changing skyscraper together, only to end
• 0:09
  up in federal court fighting over whether it should be a
• 0:11
  Public Library or a luxury condo.
• 0:14
  That's exactly what's happening today as the architects of the
• 0:17
  AI revolution face off. The skyscraper is worth $130
• 0:21
  billion, and the verdict could redefine the entire tech
• 0:25
  industry. Yeah, the stakes are just, I
• 0:27
  mean, they're astronomical, right?
• 0:28
  Now, they really are. So for today's Deep Dive, we're
• 0:31
  navigating the massive lawsuit that is completely reshaping AI,
• 0:35
  and we'll explore how government defense contracts are causing
• 0:38
  internal tech rivals. Right.
• 0:39
  Plus, we're looking at local AI automation, testing a vintage
• 0:42
  1930s AI model, and going through a blitz of quick hits.
• 0:46
  Exactly. And then we're finishing up with
• 0:48
  a really cool life saving community workflow.
• 0:51
  It is a packed show for you today.
• 0:53
  It really is. I mean, we're watching the rules
• 0:55
  of the entire industry change in real time.
• 0:57
  Totally. So let's start with this
• 0:58
  lawsuit. Elon Musk's $130 billion trial
• 1:01
  against Open AI has officially kicked off in federal court.
• 1:05
  Yeah, And just the imagery of this is wild.
• 1:07
  You have Musk on the witness stand, Right, Right.
• 1:09
  And Sam Altman is just sitting there in the gallery, watching
• 1:12
  him. Just staring him down.
• 1:13
  And you know, Musk isn't just asking for the 130 billion in
• 1:17
  damages. He wants Altman and Greg
• 1:18
  Brockman removed from the board entirely.
• 1:21
  Right. And he's also trying to force an
• 1:23
  unwind of open AI's recent for profit conversion, which is, I
• 1:29
  mean, that's a huge structural threat to the company.
• 1:31
  Huge. And during his testimony, Musk
• 1:33
  said something pretty wild. He said if a verdict comes out
• 1:37
  that it's OK to loot a charity, the entire foundation of
• 1:40
  charitable giving in America will be damaged.
• 1:43
  That's a strong phrase. Loot a charity.
• 1:45
  OK, let's unpack this because, you know, calling it looting a
• 1:49
  charity is intense, but I kind of get his angle.
• 1:51
  It's almost like a really messy divorce.
• 1:55
  Like, imagine you give someone a modest engagement ring.
• 1:57
  You break up, and then years later that ring turns out to be
• 2:00
  a priceless historical artifact. Oh, right now you want it back.
• 2:04
  Or at least you're furious about how much it appreciated in value
• 2:07
  without you. Right, because Open AI started
• 2:10
  as a nonprofit, so seeing it turn into this massive
• 2:13
  commercial juggernaut is a tough pill to swallow.
• 2:15
  If you walked. Away.
• 2:16
  Exactly. But Open AI's defense is
• 2:18
  basically calling this whole thing sour grapes.
• 2:20
  Yeah, they're arguing that Musk just didn't like seeing their
• 2:23
  massive success after he left, which, you know, is a compelling
• 2:27
  argument for a jury. And then Microsoft's legal team
• 2:29
  jumps in. Oh, Microsoft's involvement is
• 2:31
  fascinating. They pointed out that Musk
• 2:34
  didn't object to open AI structure until after its
• 2:37
  success as a direct competitor to his own company, XAI.
• 2:42
  Right. It's like, oh, now that you have
• 2:43
  a rival company, suddenly you have a legal problem with our
• 2:46
  structure. Exactly.
• 2:48
  Plus Microsoft is claiming they knew absolutely nothing about
• 2:52
  Sam Altman's brief firing in 2023.
• 2:55
  They're really trying to distance themselves.
• 2:57
  So where does this leave us? But if we connect this to the
• 3:00
  bigger picture, it's not just corporate drama.
• 3:03
  This is a fundamental legal test of what happens when a nonprofit
• 3:07
  structure incubates A multibillion dollar commercial
• 3:10
  juggernaut. Because the laws just weren't
• 3:12
  built for that right? Exactly, and the craziest part
• 3:15
  is that hundreds of pages of private emails are about to go
• 3:18
  public, so we are about to get massive industry transparency
• 3:23
  over the next 4 weeks. And the discovery process on
• 3:26
  this is going to be wild. But Speaking of massive fights
• 3:29
  inside these companies, let's transition a bit, because while
• 3:32
  they're fighting over the financial structure, there's a
• 3:34
  huge battle over who actually gets to use the AI you're.
• 3:37
  Talking about the military contracts.
• 3:38
  Exactly. So Google just signed a
• 3:40
  classified AI deal with the Pentagon, and the wording opens
• 3:44
  its models to any lawful government purpose without
• 3:47
  Google having a legal right to veto how the Pentagon uses it.
• 3:51
  Which is a huge shift for them, right?
• 3:53
  And immediately, over 600 Google staffers wrote an open letter to
• 3:57
  CEO Sundar Pichai protesting this, asking him to refuse
• 4:00
  classified workloads. Yeah, and context is important
• 4:03
  here for you guys listening. Open AI and XAI inked similar
• 4:06
  Pentagon deals last month, so it's an industry trend.
• 4:09
  But Anthropic is doing the opposite.
• 4:11
  Exactly, Anthropic is currently fighting in court after being
• 4:14
  blacklisted because they refused to drop their safety guardrails.
• 4:18
  Wow, and you know historically Google had a no weapons pledge
• 4:22
  in 2018 after successful staffed protests.
• 4:25
  Right, but that pledge was officially squad from it's AI
• 4:28
  principles in 2025. Which brings up a big question,
• 4:32
  how does a company manage this PR and internal nightmare?
• 4:36
  Like you have over 600 employees, your top talent
• 4:38
  revolting against leadership, but you're also trying to
• 4:41
  compete for massive government contracts.
• 4:44
  It's incredibly difficult. And look, maintaining neutrality
• 4:47
  here, the tech industry is actively crossing a Rubicon.
• 4:50
  What do you? Mean well, Look at the contrast.
• 4:51
  You have Anthropic clinging to guardrails and getting
• 4:54
  blacklisted versus Google's scrubbing its 2018 pledge just
• 4:58
  to secure the bag. The backlash the ChatGPT face
• 5:01
  might soon arrive at Gemini's doorstep.
• 5:03
  It's just a collision of ethics and massive financial
• 5:05
  incentives. Totally.
• 5:06
  Well, while the government is using AI for classified
• 5:08
  operations, AI is also gaining unprecedented control over our
• 5:13
  local devices. Yes.
• 5:15
  The Codex update. Right, so a new guide details
• 5:18
  how Open AI Codex can automate repetitive manual tasks using
• 5:22
  this computer use feature. This is a huge deal for everyday
• 5:26
  workflows. O you open codecs, enable the
• 5:29
  computer use plugin and then you switch permissions from default
• 5:33
  to full access. Right, and that full access part
• 5:36
  is key. Here's where it gets really
• 5:38
  interesting. Giving codecs full access is
• 5:41
  basically handing the steering wheel of your computer over to
• 5:44
  an eager, hyper competent intern who can just click through your
• 5:48
  most annoying chores. I love that analogy, it's so
• 5:51
  true. Like you can prompt it to open
• 5:54
  Chrome, debug the UI of localhost .3000, reproduce the
• 5:59
  bug and ask before making changes and it just.
• 6:01
  Does it? And it works for local labs too
• 6:02
  right? Like Photoshop exports or
• 6:04
  Premier cleanup? Yeah, file renaming, all of it.
• 6:06
  It's wild. Well, this marks a major shift
• 6:08
  from AI. AI as a text generator to an
• 6:11
  agentic operator. An agentic operator.
• 6:13
  I like that. Yeah, because instead of relying
• 6:15
  on back end API's, the AI is natively operating software
• 6:19
  designed for humans. It's visually seeing Photoshop
• 6:22
  and clicking the buttons. It's using our tools, but you
• 6:25
  know, all these modern capabilities rely on models
• 6:28
  trained on the modern Internet. Which made me wonder what
• 6:31
  happens if you completely erase the modern world from an AI's
• 6:35
  brain? Oh, this is the talkie
• 6:36
  experiment. This is fascinating, yes.
• 6:38
  So researchers Nick Levine, David Duvana, who used to be at
• 6:41
  Anthropic, and Alec Radford from Open AI, they demoed this 13
• 6:46
  billion parameter model called Talkie.
• 6:48
  And the training data is what makes it special.
• 6:50
  Exactly. Talkie is trained entirely on
• 6:53
  260 billion tokens of pre 1931 public domain data.
• 6:58
  Just books, newspapers, patents. No modern Internet at all.
• 7:02
  That's incredible. So how did they get it to act
• 7:04
  like a modern chatbot without modern chat data?
• 7:07
  OK this is so funny. They instructed it using 1930s
• 7:10
  etiquette manuals and cookbooks, and then Claude Sonnet 4.6
• 7:14
  created its answers to teach it how to chat.
• 7:16
  That's brilliant, just using vintage manners to teach
• 7:19
  conversational AI. Right.
• 7:20
  But here's the crazy twist. Python, the coding language
• 7:23
  didn't exist in 1930. Obviously.
• 7:25
  Not, but talkie still wrote working Python code.
• 7:28
  Wait, really? By generalizing logic like it
• 7:32
  flipped A+ sign to a minus sign in an example just by
• 7:35
  understanding the underlying logic.
• 7:37
  It's literally a time traveling AI figuring out modern code.
• 7:41
  What's fascinating here is how this completely sidesteps the
• 7:44
  issue of benchmark poisoning. Oh, right, explain benchmark
• 7:47
  poisoning really quick. So benchmark poisoning is when
• 7:50
  an AI model accidentally memorizes the test data because
• 7:54
  it was scraped from the modern web.
• 7:56
  So we don't know if it's actually smart or just cheating.
• 7:58
  Exactly. But since Talkie has never seen
• 8:00
  the modern web, it's ability to code in Python is the ultimate
• 8:05
  aha moment. It proves that large language
• 8:07
  models are actually learning underlying reasoning and logic,
• 8:11
  not just memorizing modern web snippets.
• 8:13
  That is mind blowing. It's actually thinking and they
• 8:15
  said a GPT 3 level version of talkie is coming next.
• 8:18
  I can't wait to see that. Me neither, but while Talkie is
• 8:21
  exploring the past, the rest of the industry is sprinting toward
• 8:24
  the future. Let's do a blitz of the day's
• 8:27
  other major news. Thanks.
• 8:28
  Do it OK? Open AI launched GPT 5.5 Codex
• 8:32
  and managed agents on Amazon Bedrock right after a contract
• 8:35
  restructure with Microsoft. And at the same time, the Wall
• 8:39
  Street Journal reported that Open AI fell short of revenue
• 8:41
  and user targets, and their CFO Sarah Fryer is questioning their
• 8:45
  massive spending. Yeah, Open AI called that report
• 8:48
  ludicrous. Well, going back to the trial we
• 8:50
  talked about, the pressure for revenue growth is the exact
• 8:54
  friction point Elon Musk is attacking in court right now.
• 8:56
  Everything is connected OK in hardware.
• 8:59
  NVIDIA released Nimotron 3 Nano Omni.
• 9:01
  It's a vision audio and text model that operates at 9 times
• 9:05
  the speed. Of Rival 9 times is a massive
• 9:07
  leap and Xiaomi open sourced MIMO V 2.5 Pro.
• 9:10
  It has a 1,000,000 context window and tied Kimi K 2.6 on
• 9:14
  the. Leaderboards The open source
• 9:16
  community is just not slowing down.
• 9:17
  And in the creative space, Spatial launched E2, beating
• 9:21
  World Labs Marble 1.1 in text and photo to 3D World
• 9:24
  Generation. Plus, Anthropic added new
• 9:26
  connectors for Blunder, Adobe Creative Cloud, Autodesk Fusion
• 9:29
  and SketchUp, and Mistral introduced workflows for
• 9:32
  enterprise agent chaining. So what does this all mean?
• 9:35
  Like we have all these massive updates, but how are real people
• 9:38
  using this? That's.
• 9:39
  The most important part right application.
• 9:42
  Totally. So we have a story from a
• 9:43
  reader, Alan C from Ontario. OK, what happened?
• 9:47
  Well, he faced a complete furnace failure on a freezing
• 9:50
  cold night. His Google NES thermostat was
• 9:53
  flashing error codes 2 and seven which are pressure switch and
• 9:56
  hardware faults and making this loud revving noise.
• 10:00
  Oh man, a furnace failing in Ontario in the cold is an actual
• 10:03
  emergency. Exactly, So instead of calling a
• 10:06
  tech, he uploaded photos of his Control Board and wiring to
• 10:09
  Google Gemini. Oh wow, using the multimodal
• 10:11
  vision. Yes.
• 10:12
  And Gemini identified a thermostat power stealing issue
• 10:16
  and a wiring phasing error. It actually guided Allen through
• 10:19
  AG to C wire conversion to provides stable power.
• 10:22
  That's a pretty complex HVAC fix for an AI to walk a novice
• 10:26
  through. Right.
• 10:27
  But within an hour, the noise stopped.
• 10:29
  He saved a $300 HVAC tech call and avoided a freezing night.
• 10:33
  That is just incredible multimodal AI acting as an on
• 10:36
  call 247 HVAC technician. Saving literal hundreds of
• 10:41
  dollars and keeping his house warm.
• 10:42
  It's amazing. It really is, and you know, it
• 10:45
  leaves us with something pretty wild to think about.
• 10:48
  What's that? Well, if an AI like Talkie
• 10:50
  trained solely on 1930s data can logically deduce how to write
• 10:56
  modern computer code, what latent undiscovered capabilities
• 11:00
  are hiding right now in the math of models we use everyday?
• 11:03
  That is a crazy thought. Right.
• 11:06
  What kind of reasoning is happening beneath the surface
• 11:08
  that we haven't even thought to test for yet?
• 11:11
  Man, we might already have alien intelligence sitting on our
• 11:14
  servers and we just haven't given it the right prompt yet.
• 11:16
  Exactly. Well, that is definitely
• 11:18
  something for you all to Mull over tonight.
• 11:21
  Thank you for joining us for this deep dive.
• 11:23
  Make sure to subscribe to stay updated on all these crazy
• 11:26
  shifts happening in AI. And hey, if you enjoyed the
• 11:29
  conversation, please rate the show 5 stars.
• 11:32
  It really helps us out. Absolutely.
• 11:33
  We'll see you back for tomorrow's deep dive.